Reducing the Dimensionality of Multivariate Time Series: An investigation of correlation structures using multivariate methods

Specialeforsvar ved Lisbeth Tomaziu

Titel:  Reducing the Dimensionality of Multivariate Time Series:
An investigation of correlation structures using multivariate methods

Abstract: This thesis is an exploratory study of the correlation structures of high dimensional, multivariate, financial time series with focus on dimensionality reduction. The purpose is to determine whether underlying mechanisms driving the correlations across periods exist by use of cluster analysis, sparse principal component analysis and vector autoregressive reduced rank regression. The results of the sparse principal component analysis performed on different time periods seemed to indicate that there were underlying drivers of the correlation structures, which had different expressions depending on time period. The reduced rank regression analysis performed on the different time periods produced connections which were harder to generalize across the periods and were interpreted as expressions of period specific behavior.

Vejledere:  Niels Richard Hansen
                  Alexander Sokol, Nordea
Censor:      Jens Ledet Jensen, Aarhus Universitet